Methodology for rapid identification of craters in fields to select optimal paths for special-purpose ground robots

This article addresses the question of how damage to agricultural land and other fields should be recognised in the context of military operations. The objective of this study is to examine the efficacy of image preprocessing techniques in facilitating object recognition. The study employed satellite images obtained from the free Google Earth Pro service (version 7.3) and processed in the specially developed proprietary software SurfaceAnalysis, which was developed in Python. The initial step involves the construction of a contrast image, which converts it from color to monochrome (black and white) while simultaneously emphasizing the most prominent texture features. In order to further enhance the visibility of objects, the application of Gaussian blurring is considered. The paper presents an analysis of the impact of changes in the standard deviation on the expressiveness of the searched elements. The present study proposes an algorithm for searching for potential objects and determining their location and boundaries for subsequent recognition based on an image obtained by Gaussian blurring. The experiments demonstrated that modifying the algorithmic parameters can markedly enhance the precision of preliminary object recognition. Consequently, it was established that preliminary digital image processing prior to recognition can expedite the search for potential objects. It is anticipated that this approach will considerably streamline the structure of the prospective neural network that will perform recognition. The research will continue to assess other preprocessing methods, refine existing algorithms, and contrast the outcomes of training neural networks on processed and unprocessed images.